About Me

Hello! My name is Cassandra (Casey) Goldberg. I am a third-year computer science PhD student at the University of Pennsylvania, advised by Eric Wong in the Brachio Lab.

My research focuses on understanding how ML models represent and use information, with the goal of making their predictions more reliable and their reasoning easier to examine. I develop mathematical theory and interpretability methods to discover concepts within learned representations, characterize when their signals are reliable, and use this understanding to evaluate and improve model behavior. Working with domain experts, I investigate how these concepts can support clinical decision-making and scientific discovery.

I graduated cum laude from Bowdoin College with a BA in computer science and physics. There, I researched statistically-sound deep learning techniques for synthetic aperture radar (SAR) image segmentation under the mentorship of Jeová Farias.

Recent News!

• Oct 2026: 🎤 I gave an invited talk on SuperActivators at SIAM NY–NJ–PA 2026.
• Oct 2026: 🎉 Our SuperActivators paper was accepted to NeurIPS 2026!
• Jul 2026: 🎤 I presented Sum-of-Checks at IPCAI 2026.
• Oct 2025: 📄 Our SAR convolutional block paper was accepted to IEEE SIBGRAPI 2025.
• May 2024: 🎓 I completed my honors thesis on SAR image segmentation at Bowdoin.
• Feb 2024: 🌟 I was selected as an AAAI-24 Undergraduate Consortium Scholar.

Publications

• SuperActivators: Only the Tail of the Distribution Contains Reliable Concept Signals
  Cassandra Goldberg, Chaehyeon Kim, Adam Stein, Eric Wong
  NeurIPS, 2026
• Sum-of-Checks: Structured Reasoning for Surgical Safety with Large Vision-Language Models
  Weiqiu You, Cassandra Goldberg, Amin Madani, Daniel A. Hashimoto, Eric Wong
  IPCAI Short Communication, 2026
• Machine Learning-Driven Analysis of Geophysical Wellbore Data for Underground Storage Applications
  Nikita Bondarenko, Ansh Ankul, Cassandra Goldberg, Sherilyn Williams-Stroud, Roman Makhnenko
  Rock Mechanics and Rock Engineering, 2026
• A Parameter Estimation-Inspired Convolutional Block for SAR Data
  Cassandra Goldberg, J. F. S. Rocha Neto
  IEEE SIBGRAPI, 2025